Data Security, Fully Baked

What does your favorite poultry dish recipe have in common with ... Department of Defense classification procedures?

While one results in a scrumptious dinner and the other in secure data, both require a complex step-by-step process. Now, with a mandate to digitize all federal records taking effect in June, experts at the University of Maryland’s Applied Research Laboratory for Intelligence and Security (ARLIS) are eating up that analogy, testing an artificial intelligence tool on 2 million recipes to see it if offers clues for simplifying classification guidelines.

As part of the Pentagon-funded project, the research team used its machine learning tool to scan recipes.com listings for commonalities, like using flour, eggs and seasoning for chicken parmesan, chicken katsu and schnitzel.

“If someone’s building the ultimate breaded chicken recipe, they can see what is inherently the same across all of those and spit that out,” says project lead Michael Brundage, an associate research engineer at ARLIS. “Then, experts can come in and debate whether that step needs to be included or not.”

As the researchers fine-tune the technology, the idea is to similarly use it to efficiently analyze the more than 1,000 classification and declassification guides across the DOD, pick out the common rules and cook up a streamlined security protocol.